A soil pollution risk prediction method, system, device and program

Abstract

The application discloses a kind of soil pollution risk prediction method, system, equipment and procedure, the method includes: obtaining monitoring data;Wherein, the monitoring data includes non-target domain historical monitoring data and target domain monitoring data;Dynamic graph attention network is constructed based on the non-target domain historical monitoring data;Cross-domain transfer learning is carried out based on the dynamic graph attention network and the target domain monitoring data, and cross-domain prediction model is obtained;Soil pollution risk prediction of target domain is carried out based on cross-domain prediction model.Through the processing scheme of the present disclosure, the problems of low prediction accuracy and poor adaptability caused by data scarcity and distribution difference in soil pollution prediction are solved.

Description

Technical Field

[0001] This invention relates to the field of soil pollution risk prediction technology, and in particular to a method, system, equipment and program for soil pollution risk prediction. Background Technology

[0002] In existing technologies, static prediction methods are typically used to predict soil pollution, or prediction models are trained based on laboratory data, etc.

[0003] Such predictive models often have certain shortcomings: 1. Limitations of static prediction: Traditional models (such as Kriging interpolation) cannot capture the spatiotemporal nonlinear evolution of pollutant diffusion; 2. Cross-regional failure: The prediction error of the model trained on laboratory data in real farmland scenarios is >52%; 3. Feature redundancy: Multi-source data (such as remote sensing images and soil sampling data) are not decoupled, resulting in a low identification rate of key pollution driving factors; 4. Passive response: The existing system can only output the risk level and cannot quantify the probability distribution of pollution in the next 72 hours.

[0004] Therefore, it is evident that the existing soil pollution prediction methods described above still have inconveniences and shortcomings in their application, and urgently need further improvement. Creating a new soil pollution prediction method has become a pressing goal for the industry. Summary of the Invention

[0005] In view of this, the present disclosure provides a method for predicting soil pollution risk, which at least partially solves the problems existing in the prior art.

[0006] In a first aspect, embodiments of this disclosure provide a method for predicting soil pollution risk, the method comprising the following steps: Acquire monitoring data; wherein, the monitoring data includes historical monitoring data of non-target domains and monitoring data of the target domain; A dynamic graph attention network is constructed based on the historical monitoring data of the non-target domain. Cross-domain transfer learning is performed based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model. Soil pollution risk prediction for the target domain is based on a cross-domain prediction model.

[0007] According to a specific implementation of an embodiment of this disclosure, the step of constructing a dynamic graph attention network based on the historical monitoring data of the non-target domain includes: The monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain are quantified as features; wherein, the monitoring points include: coordinates, timestamp, and sensor type; the environmental parameters include: pH value, rainfall, soil permeability coefficient, and groundwater flow velocity; the pollution source attributes include: pollution source type, emission amount, chemical composition, and emission time pattern; Align the node features of the quantified monitoring points, environmental parameters, and pollution source attributes to generate a dynamic graph attention network.

[0008] According to a specific implementation of an embodiment of this disclosure, the step of performing cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model includes: The general features of the dynamic graph attention network are learned using the MAML algorithm to obtain meta-parameters; Based on the meta-parameters and the target domain monitoring data, adversarial training is performed to obtain a cross-domain prediction model.

[0009] According to a specific implementation of this disclosure, the meta-parameters are obtained by learning the general features of the dynamic graph attention network through the MAML algorithm, and are derived using the following formula: ; in, These are meta-parameters; The optimization operator indicates the search for parameters that minimize the objective function. ; The loss function for a single task; The learning rate for the inner loop; For parameters The gradient; T is the number of meta-tasks; This represents the total number of non-target domains. Loss function for a single task It can be obtained through the following formula: ; in, The target domain node concentration predicted by the k-th Monte Carlo Dropout; The actual concentration of the target domain node; N is the total number of samples; For sample index, =1,2,...,N.

[0010] According to a specific implementation of this disclosure, the step of performing adversarial training based on the meta-parameters and the target domain monitoring data to obtain a cross-domain prediction model includes: S1: Adjust the discriminator based on meta-parameters, including: keeping the meta-parameters unchanged, the discriminator learns to distinguish between meta-parameter features and target domain monitoring data features; adjust the discriminator based on the recognition accuracy of meta-parameter features to maximize adversarial loss; S2: Update meta-parameters based on the adjusted discriminator, including: adjusting meta-parameters by backpropagating gradients based on the discriminator's distinguishing features and the features of the target domain monitoring data, so as to minimize the adversarial loss; Repeat steps S1 to S2 until the discriminator's prediction accuracy for both the non-target domain and the target domain reaches a preset threshold, and then output the cross-domain prediction model.

[0011] According to a specific implementation of this disclosure, the adversarial loss is obtained by the following formula: ; in, To counteract losses, a smaller value indicates a smaller difference in the domain; For discriminator networks; Samples that are not in the target domain; For the target domain sample; These are hidden layer features that are not part of the target domain; These are the hidden layer features of the target domain; Secondly, embodiments of this disclosure provide a soil pollution risk prediction system, the system comprising: The data acquisition module is configured to acquire monitoring data; wherein the monitoring data includes historical monitoring data of non-target domains and monitoring data of the target domain; The model building module is configured to build a dynamic graph attention network based on the non-target domain historical monitoring data; and to perform cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model. The prediction module is configured to predict soil pollution risk in a target domain based on a cross-domain prediction model.

[0012] According to a specific implementation of an embodiment of this disclosure, the system further includes: The dynamic graph attention network construction module is configured to perform feature quantification on the monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain; wherein, the monitoring points include: coordinates, timestamp, and sensor type; the environmental parameters include: pH value, rainfall, soil permeability coefficient, and groundwater flow velocity; the pollution source attributes include: pollution source type, emission amount, chemical composition, and emission time pattern; Furthermore, the node features of the quantified monitoring points, environmental parameters, and pollution source attributes are aligned to generate a dynamic graph attention network.

[0013] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, cause the at least one processor to perform the soil pollution risk prediction method as described in the first aspect or any implementation thereof.

[0014] Fourthly, embodiments of this disclosure also provide a non-transitory computer-readable storage medium storing computer instructions that, when executed by at least one processor, cause the at least one processor to perform the soil pollution risk prediction method in the first aspect or any implementation thereof.

[0015] Fifthly, embodiments of this disclosure also provide a computer program product, the computer program product including a computing program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the soil pollution risk prediction method in the first aspect or any implementation thereof.

[0016] The soil pollution risk prediction method in this disclosure eliminates the distribution differences between the target and non-target domains through adversarial training or feature alignment techniques, enabling the model to directly transfer and predict without additional labeling of target domain data. Fine-tuning using a small amount of labeled or unlabeled target domain data significantly reduces the cost of field sampling. For example, it maintains high prediction accuracy even when only 10% of the samples in the target domain are labeled. An integrated online learning mechanism dynamically updates model parameters to respond to sudden pollution events (such as industrial leaks). It integrates multi-dimensional information such as soil physicochemical properties (pH, organic matter), meteorological data (rainfall, wind speed), and topographic features (slope, altitude) to improve prediction accuracy. This invention solves the problems caused by data scarcity and distribution differences in soil pollution prediction through cross-domain transfer learning technology, achieving breakthroughs in prediction accuracy, adaptability, and interpretability, and possesses significant engineering application value and promotion potential. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a soil pollution risk prediction method provided in an embodiment of the present disclosure; Figure 2 A flowchart illustrating a soil pollution risk prediction method provided in this embodiment of the present disclosure; Figure 3This is a schematic diagram of the structure of a soil pollution risk prediction system provided in an embodiment of the present disclosure; Figure 4 A schematic diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0018] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0019] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0020] It should be noted that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. Furthermore, this device and / or this method can be implemented using other structures and / or functionalities besides one or more of the aspects set forth herein.

[0021] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0022] This invention provides a method for predicting soil pollution risk. It constructs a dynamic heterogeneous graph that integrates monitoring sites, environmental parameters, and pollution source attributes; dynamically calculates edge weights based on pollutant migration rates and spatial correlation; captures periodic fluctuations in pollutant concentration time series (e.g., the impact of seasonal rainfall) using a Transformer encoder; aggregates neighborhood node features in the spatial dimension using a Graph Attention Network (GAT) and identifies key diffusion paths through multi-head attention. The model is trained in non-target domains to learn pollution diffusion patterns, and the MAML algorithm is used to quickly adapt it to target areas (e.g., farmland).

[0023] Figure 1This is a schematic diagram of the soil pollution risk prediction method provided in the embodiments of this disclosure.

[0024] Figure 2 To and Figure 1 The corresponding flowchart of the soil pollution risk prediction method.

[0025] like Figure 1 As shown, in step S110, monitoring data is acquired; wherein, the monitoring data includes historical monitoring data of non-target domain and monitoring data of target domain.

[0026] In this embodiment of the invention, the monitoring point includes: coordinates (latitude and longitude), timestamp, and sensor type (e.g., pH sensor, heavy metal detector); the environmental parameters include: pH value, rainfall, soil permeability coefficient, groundwater flow velocity, and soil moisture content; the pollution source attributes include: pollution source type (industrial / agricultural), emission amount, and chemical composition (e.g., Cd). 2+ Pb 2+ The above monitoring data, including emission time patterns, can all be used as nodes.

[0027] More specifically, the monitoring data acquired includes: collecting hourly data on pH, heavy metal concentration (accuracy ±0.1 mg / kg), and soil moisture content of the soil under test using soil sensors; collecting real-time 1km×1km gridded rainfall prediction data using weather radar; and obtaining soil stratification permeability coefficients (0-2m depth resolution) using geological borehole data.

[0028] Furthermore, in this embodiment of the invention, the method further includes processing the monitoring data: First, construct pollution diffusion sensitive factors:

[0029] in, This is a time-dependent pollution diffusion sensitivity factor (or diffusion coefficient); the larger the value, the faster the diffusion. The permeability coefficient is time-dependent and is affected by rainfall erosion. Soil moisture content is time-dependent; Time-dependent rainfall erosion rate (unit: kg / (m²)) 2 ·s)); Finally, the ST-Transformer model was used to align the temporal resolution (1 hour) and spatial resolution (100m×100m) of meteorological data and soil monitoring data.

[0030] More specifically, we now proceed to step S120.

[0031] In step S120, a dynamic graph attention network is constructed based on the historical monitoring data of the non-target domain.

[0032] The dynamic graph attention network is rebuilt every 2 hours, and the adjacency relationship is updated based on the real-time penetration coefficient.

[0033] In this embodiment of the invention, the construction of a Dynamic Graph Attention Network (DGAT) based on the historical monitoring data of the non-target domain includes: quantizing the monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain; aligning the node features of the quantized monitoring points, environmental parameters, and pollution source attributes to generate a dynamic graph attention network.

[0034] More specifically, feature quantization is performed based on the following methods: Feature quantification of monitoring points is achieved through embedding encoding, such as: geographic coordinates → Geohash encoding, timestamps → periodic embedding (e.g., Sinusoidal).

[0035] Environmental parameters are quantified by normalization and time-series feature extraction. For example, historical trends can be extracted by LSTM and periodicity (such as daily / seasonal changes) can be captured by Fourier transform.

[0036] The characteristics of pollution sources can be quantified by physical models. For example, emissions can be quantified by Fick's law diffusion coefficient, and chemical components can be characterized by toxic equivalent (TEQ).

[0037] Furthermore, multi-head cross-attention is used to align node features:

[0038] in, For nodes after fusing multi-source features The final embedded representation; For nodes The query vector; For nodes The key vector; For nodes The value vector; the fused embedding vector Characteristics of the initial node in the GAT network.

[0039] Different modal nodes (such as sensor data and environmental parameters) interact through cross-attention to generate a unified-dimensional embedding vector (such as 256-dimensional).

[0040] Then, the edge weights are dynamically calculated: 1. Define edge connection rules Spatial proximity: Spatially distributed nodes (such as geographic coordinates, sensor locations) are triangulated based on Delaunay triangulation. If the nodes... and Euclidean distance (For example, if r=500m is preset), then connect the nodes to generate an initial adjacency network, ensuring that geographically adjacent nodes are connected; Semantic relevance: Screening strongly correlated node pairs (such as pollution sources and downstream monitoring points) through mutual information. Dynamic threshold: The number of edge connections is adaptively adjusted over time to avoid overfitting (e.g., using a sliding window to retain the first 50% of adjacent edges).

[0041] 2. Edge weight formula edge weight It consists of two parts: physical drive and data drive.

[0042] in, Let be the edge weight from node i to node j; Attention coefficient; The physical driving term for edge weights; This is the data-driven term for edge weights. Semantic relevance is used to adjust the edge weights based on mutual information calculation results.

[0043] Physics-driven items Calculated based on the following formula (Fick's Law and semivariogram):

[0044] in, The time-dependent diffusion coefficient, The attenuation factor is preferably 0.05; For time; It is a constant; For nodes Historical characteristic sequence; For nodes Historical characteristic sequence; Data-driven items Calculated based on the following formula (based on the MMD kernel function):

[0045] in, For nodes In time Features; For nodes In time The characteristics; MMD is a kernel function that quantifies the difference in feature distribution, used to measure the node's characteristics. , The characteristic distribution differences; This is a bandwidth parameter used to control the sensitivity to distribution differences, and is preferably 1.0.

[0046] Based on node features and edge weights A time-dependent dynamic graph attention network is generated, and the constructed graph attention network contains the spatial causal relationship of pollution propagation. By capturing the nonlinear evolution of complex environmental systems through the dynamic graph attention network, the core problem that traditional static models cannot adapt to the spatiotemporal variability of pollutant diffusion is solved.

[0047] Next, proceed to step S130.

[0048] In step S130, cross-domain transfer learning is performed based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model.

[0049] In this embodiment of the invention, the step of performing cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model includes: learning the general features of the dynamic graph attention network through the MAML algorithm to obtain meta-parameters; and performing adversarial training based on the meta-parameters and the target domain monitoring data to obtain a cross-domain prediction model.

[0050] For example, the target domain monitoring data may include: soil pH value, organic matter content, soil permeability coefficient, rainfall intensity, etc.

[0051] In this embodiment of the invention, the meta-parameters are obtained by learning the general features of the dynamic graph attention network through the MAML algorithm, and are derived using the following formula: ; in, These are the meta-parameters, the initial model parameters adapted to the target domain, serving as the starting point for adversarial training; These are the parameters for the current task; The optimization operator indicates the search for parameters that minimize the objective function. ; The loss function for a single task; The learning rate for the inner loop; For parameters The gradient; T is the number of meta-tasks; This represents the total number of non-target domains. In this embodiment of the invention, the above data can be set as follows: The dimension is 256, T=10, ∈[0, 0.5]、 =0.01、 ∈[0.1, 1.0].

[0052] Furthermore, the MAML algorithm will be described in detail below through examples.

[0053] First, the physical properties of non-target domains (such as the relationship between permeability coefficient and rainfall intensity) are encoded as initialization model parameters. Find a set of parameters through meta-training This allows it to adapt quickly with a small amount of target domain data, and when target domain data is scarce, it only requires a small number of gradient updates to adapt.

[0054] Secondly, calculate the loss function for a single task. It can be obtained through the following formula: ; in, The target domain node concentration predicted by the k-th Monte Carlo Dropout; For target domain nodes The true concentration; For sample index, =1,2,...,N; N is the total number of samples, which includes the number of monitoring points and / or the total number of data samples within a fixed time period, N>1; Inner loop parameters updated, temporary parameters generated:

[0055] in, These are the updated temporary parameters; The parameters for the current iteration; The learning rate for the inner loop; For parameters The gradient; The loss function for a single task.

[0056] Calculate the objective function (assuming a total of 10 non-target domains):

[0057] Minimize the loss of all tasks after the update, Learn the general characteristics across tasks.

[0058] Final output:

[0059] In this embodiment of the invention, the step of performing adversarial training based on the meta-parameters and the target domain monitoring data to obtain a cross-domain prediction model includes: S1: adjusting the discriminator based on the meta-parameters, including: keeping the meta-parameters unchanged, the discriminator learning to distinguish between meta-parameter features and target domain monitoring data features; adjusting the discriminator based on the recognition accuracy of the meta-parameter features to maximize the adversarial loss; S2: updating the meta-parameters based on the adjusted discriminator, including: adjusting the meta-parameters based on the discriminator's ability to distinguish between meta-parameter features and target domain monitoring data features through gradient backpropagation to minimize the adversarial loss; Repeat steps S1 to S2 until the discriminator's prediction accuracy for the non-target domain and the target domain reaches a preset threshold, and output the cross-domain prediction model. In this embodiment of the invention, the preset threshold is 55%.

[0060] In this embodiment of the invention, the adversarial loss is obtained by the following formula:

[0061] in, To counteract losses, a smaller value indicates a smaller difference in the domain; For discriminator networks; Samples that are not in the target domain; For the target domain sample; These are hidden layer features that are not part of the target domain; These are the hidden layer features of the target domain; More specifically, for example, in the meta-training phase, a meta-model is trained in 10 non-target domains, and the model parameters are optimized for each task through 5 steps of gradient updates. Loss function:

[0062] in, The loss function; The loss function for a single task; For current task parameters, These are meta-parameters; This is the regularization parameter, preferably 0.01.

[0063] Domain adaptation phase: When fine-tuning the target region, an adversarial loss is introduced:

[0064] Next, proceed to step S140.

[0065] In step S140, the soil pollution risk of the target domain is predicted based on the cross-domain prediction model.

[0066] This invention has the following advantages: 1. Strong cross-domain adaptability By using adversarial training or feature alignment techniques, the distributional differences between the non-target domain (laboratory / historical data / data from other regions) and the target domain (field monitoring) can be eliminated, enabling the model to directly transfer predictions without the need for additional labeling of target domain data.

[0067] 2. Efficiently address the problem of data scarcity Fine-tuning using a small amount of labeled or unlabeled data in the target domain significantly reduces the cost of in-situ sampling. For example, high prediction accuracy can still be maintained even when only 10% of the samples in the target domain are labeled.

[0068] 3. Real-time prediction capability It integrates an online learning mechanism, which can dynamically update model parameters to respond to sudden pollution events (such as industrial leaks).

[0069] 4. Multimodal feature fusion By integrating multi-dimensional information such as soil physicochemical properties (pH, organic matter), meteorological data (rainfall, wind speed), and topographic features (slope, altitude), the accuracy of predictions can be improved.

[0070] This invention solves the problems caused by data scarcity and distribution differences in soil pollution prediction through cross-domain transfer learning technology. It achieves breakthroughs in prediction accuracy, adaptability, and interpretability, and has significant engineering application value and promotion potential.

[0071] Figure 3 The present invention illustrates a soil pollution risk prediction system 300, which includes a data acquisition module 310, a model building module 320, and a prediction module 330.

[0072] The data acquisition module 310 is used to acquire monitoring data; wherein, the monitoring data includes historical monitoring data of non-target domains and monitoring data of the target domain; The model building module 320 is used to build a dynamic graph attention network based on the non-target domain historical monitoring data; and to perform cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model. The prediction module 330 is used to predict the soil pollution risk of the target domain based on the cross-domain prediction model.

[0073] In this embodiment of the invention, the system further includes: a dynamic graph attention network construction module, configured to perform feature quantification on the monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain; wherein, the monitoring points include: coordinates, timestamp, and sensor type of the monitoring points; the environmental parameters include: pH value, rainfall, soil permeability coefficient, and groundwater flow velocity; the pollution source attributes include: pollution source type, emission amount, chemical composition, and emission time pattern; and, the system aligns the node features of the feature-quantified monitoring points, environmental parameters, and pollution source attributes to generate a dynamic graph attention network.

[0074] See Figure 4 This disclosure also provides an electronic device 40, which includes: At least one processor; and, The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the soil pollution risk prediction method in the foregoing method embodiments.

[0075] This disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the soil pollution risk prediction method in the foregoing method embodiments.

[0076] This disclosure also provides a computer program product, which includes a computing program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions that, when executed by a computer, cause the computer to perform the soil pollution risk prediction method in the foregoing method embodiments.

[0077] The following is for reference. Figure 4 The diagram illustrates a structural schematic of an electronic device 40 suitable for implementing embodiments of the present disclosure. The electronic devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0078] like Figure 4As shown, electronic device 40 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from storage device 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of electronic device 40. The processing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0079] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic device 40 to communicate wirelessly or wiredly with other devices to exchange data. Although electronic device 40 with various devices is shown in the figure, it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0080] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a storage device 308, or installed from a ROM 302. When the computer program is executed by the processing device 301, it performs the functions defined in the methods of embodiments of this disclosure.

[0081] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0082] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0083] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: acquire at least two Internet Protocol (IP) addresses; send a node evaluation request including the at least two IP addresses to a node evaluation device, wherein the node evaluation device selects an IP address from the at least two IP addresses and returns it; and receive the IP address returned by the node evaluation device; wherein the acquired IP address indicates an edge node in a content delivery network.

[0084] Alternatively, the aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: receive a node evaluation request including at least two Internet Protocol (IP) addresses; select an IP address from the at least two IP addresses; and return the selected IP address; wherein the received IP address indicates an edge node in the content delivery network.

[0085] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0086] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0087] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".

[0088] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0089] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for predicting soil pollution risk, characterized in that, The method includes the following steps: Acquire monitoring data; wherein, the monitoring data includes historical monitoring data of non-target domains and monitoring data of the target domain; the monitoring data of the target domain includes: soil pH value, organic matter content, soil permeability coefficient, and rainfall intensity; A dynamic graph attention network is constructed based on the historical monitoring data of the non-target domain. Cross-domain transfer learning is performed based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model. Soil pollution risk prediction for the target domain based on a cross-domain prediction model; The construction of a dynamic graph attention network based on the historical monitoring data of the non-target domain includes: The monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain are quantified as features; wherein, the monitoring points include: coordinates, timestamp, and sensor type; the environmental parameters include: pH value, rainfall, soil permeability coefficient, and groundwater flow velocity; the pollution source attributes include: pollution source type, emission amount, chemical composition, and emission time pattern; Align the node features of the quantified monitoring points, environmental parameters, and pollution source attributes to generate a dynamic graph attention network; The method of performing cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model includes: The general features of the dynamic graph attention network are learned using the MAML algorithm to obtain meta-parameters; Based on the meta-parameters and the target domain monitoring data, adversarial training is performed to obtain a cross-domain prediction model; The general features of the dynamic graph attention network are learned through the MAML algorithm to obtain meta-parameters, which are then obtained using the following formula: ； in, These are meta-parameters; The optimization operator indicates the search for parameters that minimize the objective function. ; The loss function for a single task; The learning rate for the inner loop; For parameters The gradient; T is the number of meta-tasks; This represents the total number of non-target domains. Loss function for a single task It can be obtained through the following formula: ； in, The target domain node concentration predicted by the k-th Monte Carlo Dropout; The actual concentration of the target domain node; N is the total number of samples; For sample index, =1,2,...,N.

2. The soil pollution risk prediction method according to claim 1, characterized in that, The process of performing adversarial training based on the meta-parameters and the target domain monitoring data to obtain a cross-domain prediction model includes: S1: Adjust the discriminator based on meta-parameters, including: keeping the meta-parameters unchanged, the discriminator learns to distinguish between meta-parameter features and target domain monitoring data features; adjust the discriminator based on the recognition accuracy of meta-parameter features to maximize adversarial loss; S2: Update meta-parameters based on the adjusted discriminator, including: adjusting meta-parameters by backpropagating gradients based on the discriminator's distinguishing features and the features of the target domain monitoring data, so as to minimize the adversarial loss; Repeat steps S1 to S2 until the discriminator's prediction accuracy for both the non-target domain and the target domain reaches a preset threshold, and then output the cross-domain prediction model.

3. The soil pollution risk prediction method according to claim 2, characterized in that, The adversarial loss is obtained using the following formula: ； in, To counteract losses, a smaller value indicates a smaller difference in the domain; For discriminator networks; Samples that are not in the target domain; For the target domain sample; These are hidden layer features that are not part of the target domain; These are the hidden layer features of the target domain.

4. A soil pollution risk prediction system, characterized in that, The system includes: The data acquisition module is configured to acquire monitoring data; wherein, the monitoring data includes historical monitoring data of non-target domains and monitoring data of the target domain; the monitoring data of the target domain includes: soil pH value, organic matter content, soil permeability coefficient, and rainfall intensity; The model building module is configured to build a dynamic graph attention network based on the non-target domain historical monitoring data; and to perform cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model. The prediction module is configured to predict soil pollution risk in a target domain based on a cross-domain prediction model. The construction of a dynamic graph attention network based on the historical monitoring data of the non-target domain includes: The monitoring points, environmental parameters, and pollution source attributes of the historical monitoring data of the non-target domain are quantified as features; wherein, the monitoring points include: coordinates, timestamp, and sensor type; the environmental parameters include: pH value, rainfall, soil permeability coefficient, and groundwater flow velocity; the pollution source attributes include: pollution source type, emission amount, chemical composition, and emission time pattern; Align the node features of the quantified monitoring points, environmental parameters, and pollution source attributes to generate a dynamic graph attention network; The method of performing cross-domain transfer learning based on the dynamic graph attention network and the target domain monitoring data to obtain a cross-domain prediction model includes: The general features of the dynamic graph attention network are learned using the MAML algorithm to obtain meta-parameters; Based on the meta-parameters and the target domain monitoring data, adversarial training is performed to obtain a cross-domain prediction model; The general features of the dynamic graph attention network are learned through the MAML algorithm to obtain meta-parameters, which are then obtained using the following formula: ； in, These are meta-parameters; The optimization operator indicates the search for parameters that minimize the objective function. ; The loss function for a single task; The learning rate for the inner loop; For parameters The gradient; T is the number of meta-tasks; This represents the total number of non-target domains. Loss function for a single task It can be obtained through the following formula: ； in, The target domain node concentration predicted by the k-th Monte Carlo Dropout; The actual concentration of the target domain node; N is the total number of samples; For sample index, =1,2,...,N.

5. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, cause the at least one processor to perform the soil pollution risk prediction method as described in any one of claims 1 to 3.

6. A computer program product, characterized in that, The computer program product includes a computing program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, cause the computer to perform the soil pollution risk prediction method as described in any one of claims 1 to 3.

Images